The problem with Nuclear power is that it looks really good in the abstract sense but where it encounters reality it does not suit society so well.

Someone on here a few days ago pointed out that with something like 400 power stations and large industrial reactors there have only been three serious containment incidents.

I would say that isnt actually that good a track record given the cost of such calamity. Which really is unbound as far we know because it remains unclear what dealing with Chernobyl may still require as there is still fissile activity producing enough heat it could destabilze the surroundings, and all kinds of weird fungus that might be attacking the structure of the buildings, and Fukushima while it looks like we have handle on the ongoing clean up requirements, there can still be surprises. If scale the numbers of 2017-19 Chevbolt fires to 30 years you still have a rate of about 2 tenths of percent where spontaneous ignition would occur. GM still did a a recall and that only burns down your garage...

The other problem is the size and acute nature of catastrophe when it does happen. Yes you can prove with statistics coal plants release more radiation and kill more people one way or another. That's fine but they tend do it over decades at a rate society can absorb those costs. Even when a tail pond lets go or something we are usually looking at the loss of one town not a 1000+ sq miles. That is difference between one these 100 year floods and 500 year floods right there, in terms of social cost.

The picture for modern reactors does not change as much as Bill gates wants us to think because humans. Sure maybe it can't meltdown, maybe it can't produce hydrogen and create explosive conditions that will scatter materials on their own. But that does not mean some nut job shouting "Death to America" won't crash a jetliner into it, or execute some other kind of attack potentially spreading contaminates over 100s of miles. Remember you don't need a blast its the fire and the smoke that pump stuff high into the atmosphere where it can rain down later.

We use Hubspot. Here's what we use it for:

- the "live chat with support person" bubble on our web application
- single funnel for incoming/outgoing support tickets & support communications with our customers

Adopting it was an old decision.

We would love to not use hubspot; the service delivered vs. the ongoing costs seem like a bad deal now.

At some point, we'll get around to migrating to something else.

I don't disagree with any of that. I see a lot of LLM and agents being thrown at problems we used to have much more tailored tools for that did at least as good a job, while consuming a tiny and by some comparisons almost immeasurably small amount of compute resources.

You're not wrong, Word 6 on a 486 with less than handful of megabytes of memory was able to perform grammar checking on word doc. Now you need 10s of gigs of primary storage, perhaps as much as 50 gigs of secondary storage between the model and the office suite. How much better a job it actually does is disputable. The experience is different, and a no-knowledge user can get started immediately. There is value there in both aspects. I don't see people who get paid access thru and employer etc to claude, gpt, etc stop using it or ignore it.

You're asking a deeper question though, how does that no-knowledge user become someone who understands the system when the system is getting bigger all the time and the short term incentives to learn anything about it get smaller smaller. I don't know. There is a lot of scifi on the subject, dated from before 2020, so thinking without being colored by recent economic interests, probably worth reading.

I love when the jackals start fighting each other..

Exactly, my top use of AI is probably, this prompt:

Please review the document below for typographical and grammatical errors and make the required edits. Do not suggest edits that simply alter style or tone. You may make clarity changes, such as splitting a compound sentence into two separate thoughts or exchanging a pronoun for the proper noun it references. Replacing a less specific nouns such as 'thing' and 'it', with a more concrete noun being referenced; as long as it will not read as overly verbose is appropriate. Finally suggest verb replacements where there is a more appropriate term of art, as and example it would be reasonable to rewrite "Next run the code." with "Next execute the code." Make all your changes with tracking enabled so I can review and accept any proposed edits.

I don't think that makes the document any less your own than having a human copy editor conduct a similar review which has been done since forever.

That nightmare scenario actually did play out not long ago near me. A woman from out of state was falsely ID'd by an AI camera of some kind and was detained for months and it wrecked her life.

My sentiment is that the process was wrong and that the officers and other parties involved in this should all be fired, and the woman compensated.

None of that is what normally happens; that is not happening "millions of times". That's why it was newsworthy when it _did_ happen.

The original comment was philosophically open ended.

I made the example in my response very specific because arguing about hypothetical generalities is usually not productive.

The difference is this: it's difficult to ascertain if someone is guilty or innocent without arresting them. YOu don't arrest people via a random statistical sampling; you arrest via a biased selector: You arrest people you have high odds of beleiving are likely to be guilty. And you arrest them because you cannot ascertain their guilt via non-destructive observation. Without arresting them, they are allowed to simply walk away while you are trying to ascertain their status.

An arrest rate of 10% of all humans would be ridiculous.

A false positive rate of 10% _among the population of people you arrest_ is acceptable.

So the applicable comparison isn't "your cell phone doesn't work 10% of the time", the applicable comparison is "when your cell phone wasn't working, the tech support team guessed correctly as to the root cause 90% of the time, but guessed incorrectly 10% of the time"

IOW, 90% success rate with a 10% FP rate amongst a problematic cohort.

>No I maintain it would be pathetic performance by LEO if they were willy nilly arresting people hoping to bag the perp.

Nobody is advocating for that.

The cops should only be arresting people they have a plausible reason to arrest. If it turns out that 10% of the time their guess was incorrect, that's acceptable. They are not performing random arrests and getting a lot hit rate. They are doing targeted arrests and getting an ultra high hit rate.

>AI is being forced inappropriately into peoples lives with no opt out *everywhere*

I agree with you, and I also don't like that, and I think stopping that via market choices and legislation, as appropriate, is the right thing to do.

In my mind that's totally separate from what I am talking about and commenting on.

That said, you can make legislation so that you're not forced to use it (or be spied on by it), but you cannot make a law that stops AI development. If nothing else, the law won't reach beyond our borders. And eventually a nation hostile to you will develop ASI and then they will impose it on you (and worse)

I've been a professional software engineer for 30 years.

AI has changed everything for software engineers.

I wrote a line of code by hand saturday (client blocks AI) and it was weird. I hadn't done it in months.

I don't have a crystal ball. I cannot say for sure whats coming.

But I would say this: if the scope of your ambition is small, e.g. the possible amount of value you can add as a business (or as an employee) is fundamentally constrained by something, then having a machine do that thing way more efficiently than you is a problem.

Machines will take an increasing share of your fixed pie.

Lots of software isn't that.

Lots of software is trying to make the pie bigger.

We're doing things now that we never would have tried. The incremental cost of writing code goes to zero. Software engineering changes. What assumptions about what you do and how you spend your time and what value you provide change when the cost of writing code goes down 10x or 100x?

For all of human history, the speed of acheiving every goal has been limited by available high-IQ-hours. At every moment in human history, there was unimaginable value not being created because it didn't make the cut compared to everything else that also needed doing - the brains capable of doing it didn't have the remaining hours to do it.

Artificial intelligence is changing the math on that. Unbounded level of available hours at what appears to unbounded level of IQ.

You are trapped in a prison made of your own lack of imagination.

(I'm not saying I'm not, by the way)

I don't understand your point. Can you explain it for me?

I assume you have never tried making an AI model of the type we're discussing (near-frontier LLM), and you have also never tried making a printing press.

Neither have I.

However, knowing a bit about how LLMs work, and knowing a bit about model training for non-LLM models, and having done a bit of word working and metal work, I assume it's actually easier now to make an AI model at home than it is to make a printing press at home.

My response included assumptions which might not hold.

If the way that police work actually worked was "we pick 100 names randomly out of the phone book, arrest them all, and 10 were innocent" ... the biggest problem would _still_ be that of a random sample you have a 90% guilty rate. That's actually not even civilization. Nobody actually trapped in a "society" like that would object to being detained and released.

So the way to think about it is, "we detain 100 people for whom we have probable cause - we believe a crime was taking place, and we believe they were proximal or involved in that crime.. and we have only a 10% false positive rate"

That's really quite good.

So for instance, if you have a riot and you arrest 100 people, and then later you find out that 90% of them are actually criminals, but 10% of them actually didn't break any law, and release that 10% - I'm fine with that. It was completely reasonable to arrest all 100 of them in the first place - they were, after all, in physical proximity to a riot, and they looked and acted enough like the 90% of actual rioters that the police (correctly) detained them.

A fine way of not getting arrested in this scenario is to stay the hell away from riots. And when "protests" become riots, _leave_.

If revolution, anarchy, and violence are the goal, fine. But then don't expect due process.

It sounds like you're saying that the printing press was just a technical enabler, sometimes in the hands of bad people. You're saying that the printing press per-se was neither good nor evil; it was the people who used the printing press for evil ends that really caused the problem.

Me too.

>Should we accept jailing 10 innocent people to catch 90 actual guilty criminals?

Yes.

This is one of those questions that seems philosophically interesting but is not interesting pragmatically.

When you jail someone, you have reason to believe they are proximate to what you believe is a crime. You remove their freedom to move/act temporarily while you ascertain the existence/severity of a crime, and if they are a party to that crime.

When you figure out that you made a mistake, you release them.

The (presumed) alternative here is patently ridiculous.

A society where 0 innocent people are arrested but 90 guilty people run around is much worse than a society where 90 guilty people are caught and 10 innocent people are temporarily detained.

The actual #1 threat to Americans is not being improperly arrested, but being victimized by known repeat criminals who are repeatedly released back into society.

Bukele shows what is possible.

think "Manhattan Project"

The USG doesn't want to be significantly behind anyone else when it comes to having ASI. That's the national security _interest_.

Our elected officials didn't know anything about atomic bombs in 1942 either. They still don't. Back then, the USG provided national security controls and oversight to one of the biggest and most successful fronteir research and engineering projects in human history.

>I'm certainly not sold on the idea

I'm not either.

The Aschenbrenner paper got into my brain and nothing else has dislodged it yet.

tl;dr:
1) if China developers acquires ASI significantly before the US does, the national security posture of the American people and homeland becomes untenable.
2) the disastrous lax security and processes within the American frontier labs make it inevitable that any AGI/ASI developments that happen there will rapidly be exfiltrated to China
3) The last time there was a national security research project this serious was the Manhattan project.

The point of nationalization isn't to keep huggingface from getting hacked or to keep self driving cars from running over strollers. Nobody who matters cares about any of that stuff. We are all expendable to them.

The point of a US national security takeover is to try and delay or prevent ASI breakthroughs being leaked to adversary countries.